Relationship Types and Contraceptive Use Within Young Adult Dating Relationships CONTEXT: Although expanding research has found that relationship characteristics can shape contraceptive use among young adults, limited research has examined how relationship characteristics intersect to form distinct types of relationships and how relationship types are linked to contraceptive use. METHODS: Data from the 2002–2005 rounds of the National Longitudinal Survey of Youth 1997 cohort were used to examine contraceptive use in 3,485 young adult dating relationships. Latent class analysis was employed to develop a typology of relationships using measures of relationship structure (duration) and quality (intimacy, commitment and conflict). Multinomial logistic regression analyses were used to estimate associations between relationship type and contraceptive use and method choice at last sex. RESULTS: Four types of relationships were identified, two shorter term and two longer term, differentiated by levels of intimacy, commitment and conflict. Young adults in longer term relationships with greater conflict and lower intimacy and commitment were less likely than those in other long-term relationships to use hormonal and dual methods versus no method (relative risk ratios, 0.6–0.7). Hormonal method use, versus no method use or condom use, was more prevalent in short-term relationships with greater intimacy and commitment and lower conflict than in other short-term relationships (1.7 and 1.9, respectively).
By Jennifer Manlove, Kate Welti, Elizabeth Wildsmith and Megan Barry Jennifer Manlove is program area director and senior research scientist, Kate Welti is research scientist, Elizabeth Wildsmith is senior research scientist and Megan Barry is research analyst, all at Child Trends, Bethesda, MD.
CONCLUSIONS: Classifying short-term relationships as “casual” or long-term ones as “serious” may ignore heterogeneity within these categories that may have implications for contraceptive use. Future qualitative research could provide a better understanding of relationship types and couples’ fertility intentions and access to and use of contraceptives. Perspectives on Sexual and Reproductive Health, 2014, 46(1):41–50, doi: 10.1363/46e0514
The field of “emerging adulthood” has identified the late teenage years and early 20s as a unique developmental stage, when many individuals are sexually active but not yet in a committed relationship, and instead are moving in and out of less committed dating relationships.1–3 As a result, emerging adulthood is also a stage characterized by high rates of unintended and nonmarital childbearing and STDs.4,5 Avoiding these outcomes requires the consistent and correct use of effective contraceptives, including the use of condoms for STD prevention, hormonal methods for pregnancy prevention and dual methods (condoms and hormonal contraceptives together) for both. However, despite recent increases in contraceptive use, many young adults fail to use effective methods, and some do not use contraceptives at all.6 An expanding research base has found that decisions about contraception are often made within the relationship dyad.7–10 Some of this research (especially research on condom use) has focused on single-item measures of relationship context, such as relationship duration, which is used as a proxy for perceived relationship seriousness.10 Other studies, focusing on adolescents9,11,12 and young adults,7,8 have found that several dimensions of relationships, as well as partner characteristics, are associated with contraceptive and condom use and consistency. While it is useful Volume 46, Number 1, March 2014
to examine the independent associations that individual relationship characteristics have with contraceptive and condom use, some research suggests that many of these characteristics coalesce into meaningful relationship categories, which, in turn, are linked to contraceptive use.7,9 In the study presented here, we sought to build on this research by using data from the National Longitudinal Survey of Youth 1997 cohort to identify the ways in which characteristics of the relationship dyad are linked to contraceptive use and method choice. Our goal was to provide a better understanding of contraceptive use within young adult dating relationships, information that is critical to policymakers and practitioners focused on improving the reproductive health of this high-risk population. BACKGROUND
Correlates of Contraceptive Use A large body of research has linked a broad range of individual and partner characteristics to contraceptive use and method choice among teenagers and young adults. At the individual level, socioeconomic characteristics (of respondents or their families), demographic characteristics (such as race or ethnicity and age) and behaviors (such as a respondent’s sexual history) have all been linked to contraceptive use. For example, lower educational attainment 41
Contraceptive Use Within Young Adult Dating Relationships
and being unemployed and not enrolled in school have been linked to reduced use of contraceptives, particularly hormonal methods.8,13 Additionally, a younger age at first sex is associated with reduced contraceptive use,14 as is already having had a child.8 At the partner level, relationship asymmetries with respect to age and race or ethnicity, low partner educational attainment, and partner disengagement from school and work have been linked to less contraceptive use and consistency.7,8,11,15,16 However, much of the variability in contraceptive use remains unexplained even once these individual and partner characteristics are taken into account. Thus, an expanding body of research has begun to examine how characteristics of the relationship dyad are linked to contraceptive use. Often drawing from life-course theory, which argues that behavior cannot be understood independently of the social relationships within which the individual is embedded,17,18 this research has linked various structural and behavioral dimensions of romantic and dating relationships to contraceptive use. The four dimensions of relationships most consistently identified are duration of the relationship and levels of intimacy, commitment and conflict.19 It is well established that condom use declines as relationship duration increases.10 The sawtooth hypothesis posits that as relationships become more serious over time and couples no longer perceive themselves to be at risk of acquiring and transmitting STDs, condom use (or consistent condom use) declines and hormonal method use increases; the pattern is then repeated in each new relationship.10,20,21 Thus, longer relationship duration is generally associated with increased hormonal method use. However, over time, this association may change. For example, some research has found that very long relationship duration (four or more years) is associated with reduced pill use.13 Also, reflecting the transition from condoms to hormonal methods in more serious relationships,21,22 high levels of emotional closeness and relationship commitment are consistently associated with reduced condom use20,23–25 and greater hormonal use10,26 among teenagers and young adults. Notably, in one study, individuals in more serious relationships, but not those in casual relationships, reported less condom use when the female partner was using hormonal methods.21 Research has also found that relationship conflict is associated with reduced condom use among young adults27 and female teenagers,9 and with reduced contraceptive use, particularly condom and dual method use, among young adults.28 Conflict in a relationship could be symptomatic of power differentials or a lack of communication, both of which have been linked to lower levels of contraceptive use.9,29
*Latent class analysis is a statistical technique for examining relationships in data by identifying a set of mutually exclusive unobserved subgroups that account for the distribution of cases occurring within a cross-tabulation of discrete variables—here, relationship characteristics (source: McCutcheon A, Latent Class Analysis, Beverly Hills, CA: Sage, 1987).
42
Interactions Between Relationship Dimensions Most of the research discussed above examined the links between individual relationship characteristics and contraceptive use. However, the context of any social relationship, including romantic relationships, is made up of multiple dimensions that are correlated with one another and that interact with each other in important ways. To date, few studies have examined how the characteristics of romantic relationships may group together to influence contraceptive use. In this article, we attempt to fill this gap by identifying underlying relationship classes, defined by the important dimensions of relationships described above, and by exploring how these classes are linked to contraceptive use and method choice within young adult dating relationships. Our work builds on two studies that have combined several relationship characteristics into indices9 or relationship typologies7 to examine their association with contraceptive use. In the first study, which used a local-area sample of adolescents in Ohio, Manning and colleagues9 grouped relationship characteristics into one index of positive attributes (such as intimate self-disclosure, passionate love and relationship salience) and one index of negative attributes (such as conflict and nonexclusivity). They found that high scores on either index were associated with reduced condom consistency. Descriptive analyses also suggested that a combination of high positive and high negative scores was associated with less condom consistency. In the second study, Kusunoki and Upchurch7 used nationally representative data on adolescent and young adult relationships to examine the association between relationship characteristics and contraceptive use at last sex. They used cluster analysis to create a relationship typology for dating and cohabiting relationships based on selfassessed relationship type (sex-only, dating, cohabiting) and three structural attributes: duration of the relationship, duration of the presexual relationship and frequency of intercourse within the relationship. They found heterogeneity in contraceptive use within self-assessed relationship type. For example, in dating relationships, frequency of sexual intercourse was inversely associated with condom use and dual method use; in more casual “hook-up” relationships, duration of the presexual relationship was positively associated with condom and hormonal method use.7 We extended the work in these studies by using longitudinal data on young adult dating relationships to examine whether comparable relationship types emerge in latent class analyses.* Building on Kusunoki and Upchurch,7 we tested whether there was heterogeneity within shorter term and longer term relationships, and assessed how heterogeneity was associated with contraceptive use and method choice at last sex. However, instead of relying solely on self-reported relationship type and structural measures of the relationship, we captured additional relationship characteristics and created classes based on measures of relationship duration, intimacy, commitment and conflict. Drawing on the work of Manning and colleagues,9 Perspectives on Sexual and Reproductive Health
we examined whether relationship types that included both positive and negative attributes emerged and whether they were associated with especially low levels of condom or contraceptive use, which might be expected given the authors’ finding of lower condom consistency among these relationships. Our analyses differentiate between method types because reasons for using a condom (which provides protection against STDs) may differ from reasons for using hormonal methods and may be unique to a given relationship.21 Focusing on only one method or combining methods into a general measure of contraceptive use may ignore this important variation.
Hypotheses Consistent with a life-course perspective and with the research by Manning et al.9 and Kusunoki and Upchurch,7 we hypothesized that dating couples would be sorted into several classes based on multiple dimensions of their relationship. Some of these classes would include both positive and negative attributes. Also, consistent with the sawtooth hypothesis, we posited that casual, short-term relationships, marked by lower intimacy and commitment, would be associated with greater condom use. By contrast, we expected that relationships with greater intimacy, commitment and duration would be associated with greater hormonal method use. Given that levels of condom use tend to be relatively high in longer term casual relationships even when the women are using hormonal methods,21 we anticipated greater dual method use in longer, casual relationships, as well as in relationships that have high levels of intimacy and commitment but are of comparatively short duration. In addition, we hypothesized that relationship classes characterized by high levels of conflict would be associated with reduced use of any method. METHODS
Data and Sample We used data from the National Longitudinal Survey of Youth 1997 cohort, a nationally representative sample of 8,984 youth aged 12–16 at baseline.30 The survey collects information on respondents’ romantic partners and sexual experiences, as well as family background and demographic characteristics. We focused on sexually active individuals in current dating relationships because our key measures of intimacy, commitment and conflict were available only for respondents in such relationships (those who were cohabiting, married or no longer with their dating partner were not asked the relevant questions). Although respondents were interviewed every year starting in 1997, we primarily used data from 2002–2005 (Rounds 6–9) because respondents were not asked about their dating partners until Round 6 and several of our measures of interest were not collected after Round 9. We also used data from previous rounds and baseline for time-invariant individual and family background control measures. In each round, participants were asked “Since [date of last interview], have you been in a dating relationship in Volume 46, Number 1, March 2014
which you thought of yourself as part of a couple?” We considered respondents to be in a dating relationship if they answered yes and if the partner was a current dating partner, as opposed to a spouse, cohabiting partner or former partner. Same-sex dating relationships and those in which the respondent had previously cohabited with the partner were excluded. Overall, 4,013 respondents reported a current dating relationship during 2002–2005. Of the remaining respondents, 553 were not in the sample during the rounds of interest, 1,876 were cohabiting or married, and 2,542 had no current dating relationships. Respondents in our sample were aged 18–26. Respondents included in the sample and those who reported no relationships reported generally similar social and demographic characteristics, suggesting that our sample is fairly representative of young adults who are unmarried and not cohabiting; the only differences between the two groups were that those reporting no current dating relationship were more likely to be male and marginally less likely to be black. On the other hand, respondents who were not included because they were cohabiting or married differed from those who were dating in several ways: They were more likely to be female and Hispanic, and they were less likely to be black, to have lived with two biological or adoptive parents at baseline and to have a parent with at least some college experience. They also had a marginally lower average age at first sex. From these data, we created a relationship-level file, in which each record represented a relationship, for a total of 4,574. We then removed 19 relationships with missing data on contraceptive use at last sex and 16 relationships in which the respondent reported using methods that did not fit into our contraceptive method typology, described below. We also removed 386 relationships in which the respondents were not having sex with their dating partner. Although a respondent could have up to four records if he or she was currently dating a different partner at each interview date, more than 85% contributed just one relationship.* Because we could not look at multiple relationships for the majority of respondents, we focused on respondents’ most recent dating relationship and dropped 538 relationships that did not meet this criterion, leaving us with one relationship per respondent. Finally, we removed 130 relationships in which respondents said they were trying to get pregnant with their partner. Our final sample consisted of 3,485 unique relationships (1,804 from female respondents and 1,681 from male respondents).
Measures 䊉 Dependent variables. Our dependent variable of interest was a four-level measure of the type of contraceptive used at last sex: no method (including withdrawal and natural
*Respondents may have contributed only one relationship because they dated one partner across multiple rounds or because they had short relationships between survey dates that were no longer current at the time of the interview.
43
Contraceptive Use Within Young Adult Dating Relationships
family planning methods, which were reported by 2% of respondents); condom only; hormonal or long-acting method only; or dual method (condom and hormonal or long-acting method). 䊉 Relationship characteristics. We created six three-level categorical relationship variables for our latent class analysis. We chose three-level variables because having all variables on the same metric helps latent class analysis models converge31 and, in our data, the three-level categorical variables provided better model fit than two-level variables. We set our category cut points on the basis of previous research, but also to maintain a relatively even distribution across levels in our sample. Relationship duration was measured by the total number of months elapsed from the date the relationship began to the date of last sex. Responses were categorized as six months or less, seven months to one year or more than one year. Also, we included a variable measuring the length of the couple’s relationship prior to the first time they had sex (the presexual relationship). The categories indicated that the couple had had sex before or during the month the relationship started, had dated for 1–3 months prior to having sex or had dated for four or more months prior to having sex. Intimacy was assessed with two questions about how close respondents felt to their partner and how much they thought their partner cared about them. Responses for each ranged from 0 to 10 (high scores indicate high caring or closeness). Because these two variables were moderately correlated (r=0.66), and because a large proportion of respondents reported very high levels of closeness and caring, we averaged these measures and combined scores into one measure indicating low intimacy (a score of 7 or less), medium intimacy (a score of 8 or 9) or high intimacy (a score of 10). Relationship commitment was based on two measures. The first was a 0–10 scale in which respondents rated the likelihood that they and their partner would be together in six months. A score of zero indicated no chance; a 10 indicated 100% likelihood. We categorized this likelihood as low (less than 6), medium (6–9) or high (10). For the second measure of commitment, we included a variable for whether the couple had discussed marriage or cohabitation; responses were categorized as talked about neither, talked about cohabitation or talked about marriage. (Respondents who reported having discussed both cohabitation and marriage were included in the marriage category.) Respondents rated the level of conflict in their relationship on a 0–10 scale. Responses skewed toward the lower end of the scale, so we categorized them as low conflict *Eighteen respondents reported another race or ethnicity; these were included with whites. †We ran a latent class analysis model that included a three-level child variable (no child, child with current partner, child with other partner), and the results were unchanged.
44
(a score of 0–2), medium conflict (3–5) or high conflict (6–10). 䊉 Control variables. We included a number of family and individual controls associated with contraceptive use. Time-invariant characteristics measured at Round 1 were gender, race or ethnicity (white, black, native-born Hispanic and foreign-born Hispanic*), whether either of the respondent’s parents had completed some college or more, and whether the respondent had lived with two biological or adoptive parents at baseline. We also included a measure of the respondent’s age at first sex. Time-varying characteristics measured at the time of last sex with the dating partner were the respondent’s age, whether the respondent had completed some college or more, whether the respondent was neither enrolled in school nor employed, the number of sexual partners in the past year (capped at 10), whether the respondent had used hormonal methods prior to the current relationship and whether the respondent had children. Fifteen percent of the sample had had a child, but only 2% had had a child with the current partner. Therefore, for class creation, having had a child was considered an individual-level control, rather than a relationship-level characteristic.† Also, we controlled for several important partner characteristics. We included a measure of whether the partner was of a different race or ethnicity than the respondent, as well as a continuous measure of age difference between the respondent and his or her partner (where a positive value indicates the partner is older). Two variables measured whether the partner had completed some college or more and whether the partner was neither enrolled in school nor employed. Ideally, these two characteristics would be measured at the time of last sex; however, the relevant questions were based on the beginning of the relationship.
Analysis We conducted a latent class analysis in Mplus to identify relationship classes.31 We tested whether gender was a significant predictor of relationship class membership and found that it was not. Therefore, we conducted the class creation on the full sample of males and females. In conducting the latent class analysis, we compared the fit indices (entropy, the Bayesian information criterion, the sample-size adjusted Bayesian information criterion and the Akaike information criterion) for one- to five-class models to identify the appropriate number of classes. We then conducted descriptive and bivariate analyses to examine the average contraceptive patterns and partner and individual characteristics across classes, which we tested for differences using t tests. In multivariate analyses, we examined the association between relationship class and contraceptive use, net of partner, family and individual characteristics. Using logistic regression, we modeled contraceptive use versus nonuse (results not shown); using multinomial logistic regression, we modeled the four-category contraceptive method type. Regressions were run in Stata 11 and incorporated Perspectives on Sexual and Reproductive Health
household clustering and probability weighting variables. We ran the regression models four times, each time using a different relationship class as the reference group, in order to compare all of the classes. All associations presented were significant at p1 year
88.0 10.5 1.5
64.3 30.7 5.0
*,** 2.9 11.9 85.3
*,**,*** 16.1 15.6 68.3
Length of presexual relationship Sex before/in first month of dating 1–3 months ≥4 months
63.0 34.6 2.4
* 56.9 39.1 4.1
*,** 30.6 38.8 30.6
*,** 30.2 42.2 27.6
Intimacy† Low Medium High
77.6 22.1 0.3
* 0.7 70.5 28.7
*,** 29.1 50.1 20.7
*,**,*** 0.1 13.0 87.0
Likelihood of being with partner in six months‡ Low 76.9 Medium 21.8 High 1.3
* 8.1 70.1 21.8
*,** 38.1 55.2 6.7
*,**,*** 0.2 4.9 94.9
Discussed cohabitation/marriage Neither Cohabitation Marriage
72.8 18.0 9.3
* 38.5 21.2 40.2
*,** 18.5 12.1 69.4
*,**,*** 5.9 5.6 88.6
Conflict§ Low Medium High
39.8 36.6 23.6
* 57.8 18.5 23.7
*,** 15.6 45.8 38.6
*,*** 59.1 20.9 20.1
100.0
100.0
100.0
100.0
Total
*Differs from short-term/casual at p
By Jennifer Manlove, Kate Welti, Elizabeth Wildsmith and Megan Barry Jennifer Manlove is program area director and senior research scientist, Kate Welti is research scientist, Elizabeth Wildsmith is senior research scientist and Megan Barry is research analyst, all at Child Trends, Bethesda, MD.
CONCLUSIONS: Classifying short-term relationships as “casual” or long-term ones as “serious” may ignore heterogeneity within these categories that may have implications for contraceptive use. Future qualitative research could provide a better understanding of relationship types and couples’ fertility intentions and access to and use of contraceptives. Perspectives on Sexual and Reproductive Health, 2014, 46(1):41–50, doi: 10.1363/46e0514
The field of “emerging adulthood” has identified the late teenage years and early 20s as a unique developmental stage, when many individuals are sexually active but not yet in a committed relationship, and instead are moving in and out of less committed dating relationships.1–3 As a result, emerging adulthood is also a stage characterized by high rates of unintended and nonmarital childbearing and STDs.4,5 Avoiding these outcomes requires the consistent and correct use of effective contraceptives, including the use of condoms for STD prevention, hormonal methods for pregnancy prevention and dual methods (condoms and hormonal contraceptives together) for both. However, despite recent increases in contraceptive use, many young adults fail to use effective methods, and some do not use contraceptives at all.6 An expanding research base has found that decisions about contraception are often made within the relationship dyad.7–10 Some of this research (especially research on condom use) has focused on single-item measures of relationship context, such as relationship duration, which is used as a proxy for perceived relationship seriousness.10 Other studies, focusing on adolescents9,11,12 and young adults,7,8 have found that several dimensions of relationships, as well as partner characteristics, are associated with contraceptive and condom use and consistency. While it is useful Volume 46, Number 1, March 2014
to examine the independent associations that individual relationship characteristics have with contraceptive and condom use, some research suggests that many of these characteristics coalesce into meaningful relationship categories, which, in turn, are linked to contraceptive use.7,9 In the study presented here, we sought to build on this research by using data from the National Longitudinal Survey of Youth 1997 cohort to identify the ways in which characteristics of the relationship dyad are linked to contraceptive use and method choice. Our goal was to provide a better understanding of contraceptive use within young adult dating relationships, information that is critical to policymakers and practitioners focused on improving the reproductive health of this high-risk population. BACKGROUND
Correlates of Contraceptive Use A large body of research has linked a broad range of individual and partner characteristics to contraceptive use and method choice among teenagers and young adults. At the individual level, socioeconomic characteristics (of respondents or their families), demographic characteristics (such as race or ethnicity and age) and behaviors (such as a respondent’s sexual history) have all been linked to contraceptive use. For example, lower educational attainment 41
Contraceptive Use Within Young Adult Dating Relationships
and being unemployed and not enrolled in school have been linked to reduced use of contraceptives, particularly hormonal methods.8,13 Additionally, a younger age at first sex is associated with reduced contraceptive use,14 as is already having had a child.8 At the partner level, relationship asymmetries with respect to age and race or ethnicity, low partner educational attainment, and partner disengagement from school and work have been linked to less contraceptive use and consistency.7,8,11,15,16 However, much of the variability in contraceptive use remains unexplained even once these individual and partner characteristics are taken into account. Thus, an expanding body of research has begun to examine how characteristics of the relationship dyad are linked to contraceptive use. Often drawing from life-course theory, which argues that behavior cannot be understood independently of the social relationships within which the individual is embedded,17,18 this research has linked various structural and behavioral dimensions of romantic and dating relationships to contraceptive use. The four dimensions of relationships most consistently identified are duration of the relationship and levels of intimacy, commitment and conflict.19 It is well established that condom use declines as relationship duration increases.10 The sawtooth hypothesis posits that as relationships become more serious over time and couples no longer perceive themselves to be at risk of acquiring and transmitting STDs, condom use (or consistent condom use) declines and hormonal method use increases; the pattern is then repeated in each new relationship.10,20,21 Thus, longer relationship duration is generally associated with increased hormonal method use. However, over time, this association may change. For example, some research has found that very long relationship duration (four or more years) is associated with reduced pill use.13 Also, reflecting the transition from condoms to hormonal methods in more serious relationships,21,22 high levels of emotional closeness and relationship commitment are consistently associated with reduced condom use20,23–25 and greater hormonal use10,26 among teenagers and young adults. Notably, in one study, individuals in more serious relationships, but not those in casual relationships, reported less condom use when the female partner was using hormonal methods.21 Research has also found that relationship conflict is associated with reduced condom use among young adults27 and female teenagers,9 and with reduced contraceptive use, particularly condom and dual method use, among young adults.28 Conflict in a relationship could be symptomatic of power differentials or a lack of communication, both of which have been linked to lower levels of contraceptive use.9,29
*Latent class analysis is a statistical technique for examining relationships in data by identifying a set of mutually exclusive unobserved subgroups that account for the distribution of cases occurring within a cross-tabulation of discrete variables—here, relationship characteristics (source: McCutcheon A, Latent Class Analysis, Beverly Hills, CA: Sage, 1987).
42
Interactions Between Relationship Dimensions Most of the research discussed above examined the links between individual relationship characteristics and contraceptive use. However, the context of any social relationship, including romantic relationships, is made up of multiple dimensions that are correlated with one another and that interact with each other in important ways. To date, few studies have examined how the characteristics of romantic relationships may group together to influence contraceptive use. In this article, we attempt to fill this gap by identifying underlying relationship classes, defined by the important dimensions of relationships described above, and by exploring how these classes are linked to contraceptive use and method choice within young adult dating relationships. Our work builds on two studies that have combined several relationship characteristics into indices9 or relationship typologies7 to examine their association with contraceptive use. In the first study, which used a local-area sample of adolescents in Ohio, Manning and colleagues9 grouped relationship characteristics into one index of positive attributes (such as intimate self-disclosure, passionate love and relationship salience) and one index of negative attributes (such as conflict and nonexclusivity). They found that high scores on either index were associated with reduced condom consistency. Descriptive analyses also suggested that a combination of high positive and high negative scores was associated with less condom consistency. In the second study, Kusunoki and Upchurch7 used nationally representative data on adolescent and young adult relationships to examine the association between relationship characteristics and contraceptive use at last sex. They used cluster analysis to create a relationship typology for dating and cohabiting relationships based on selfassessed relationship type (sex-only, dating, cohabiting) and three structural attributes: duration of the relationship, duration of the presexual relationship and frequency of intercourse within the relationship. They found heterogeneity in contraceptive use within self-assessed relationship type. For example, in dating relationships, frequency of sexual intercourse was inversely associated with condom use and dual method use; in more casual “hook-up” relationships, duration of the presexual relationship was positively associated with condom and hormonal method use.7 We extended the work in these studies by using longitudinal data on young adult dating relationships to examine whether comparable relationship types emerge in latent class analyses.* Building on Kusunoki and Upchurch,7 we tested whether there was heterogeneity within shorter term and longer term relationships, and assessed how heterogeneity was associated with contraceptive use and method choice at last sex. However, instead of relying solely on self-reported relationship type and structural measures of the relationship, we captured additional relationship characteristics and created classes based on measures of relationship duration, intimacy, commitment and conflict. Drawing on the work of Manning and colleagues,9 Perspectives on Sexual and Reproductive Health
we examined whether relationship types that included both positive and negative attributes emerged and whether they were associated with especially low levels of condom or contraceptive use, which might be expected given the authors’ finding of lower condom consistency among these relationships. Our analyses differentiate between method types because reasons for using a condom (which provides protection against STDs) may differ from reasons for using hormonal methods and may be unique to a given relationship.21 Focusing on only one method or combining methods into a general measure of contraceptive use may ignore this important variation.
Hypotheses Consistent with a life-course perspective and with the research by Manning et al.9 and Kusunoki and Upchurch,7 we hypothesized that dating couples would be sorted into several classes based on multiple dimensions of their relationship. Some of these classes would include both positive and negative attributes. Also, consistent with the sawtooth hypothesis, we posited that casual, short-term relationships, marked by lower intimacy and commitment, would be associated with greater condom use. By contrast, we expected that relationships with greater intimacy, commitment and duration would be associated with greater hormonal method use. Given that levels of condom use tend to be relatively high in longer term casual relationships even when the women are using hormonal methods,21 we anticipated greater dual method use in longer, casual relationships, as well as in relationships that have high levels of intimacy and commitment but are of comparatively short duration. In addition, we hypothesized that relationship classes characterized by high levels of conflict would be associated with reduced use of any method. METHODS
Data and Sample We used data from the National Longitudinal Survey of Youth 1997 cohort, a nationally representative sample of 8,984 youth aged 12–16 at baseline.30 The survey collects information on respondents’ romantic partners and sexual experiences, as well as family background and demographic characteristics. We focused on sexually active individuals in current dating relationships because our key measures of intimacy, commitment and conflict were available only for respondents in such relationships (those who were cohabiting, married or no longer with their dating partner were not asked the relevant questions). Although respondents were interviewed every year starting in 1997, we primarily used data from 2002–2005 (Rounds 6–9) because respondents were not asked about their dating partners until Round 6 and several of our measures of interest were not collected after Round 9. We also used data from previous rounds and baseline for time-invariant individual and family background control measures. In each round, participants were asked “Since [date of last interview], have you been in a dating relationship in Volume 46, Number 1, March 2014
which you thought of yourself as part of a couple?” We considered respondents to be in a dating relationship if they answered yes and if the partner was a current dating partner, as opposed to a spouse, cohabiting partner or former partner. Same-sex dating relationships and those in which the respondent had previously cohabited with the partner were excluded. Overall, 4,013 respondents reported a current dating relationship during 2002–2005. Of the remaining respondents, 553 were not in the sample during the rounds of interest, 1,876 were cohabiting or married, and 2,542 had no current dating relationships. Respondents in our sample were aged 18–26. Respondents included in the sample and those who reported no relationships reported generally similar social and demographic characteristics, suggesting that our sample is fairly representative of young adults who are unmarried and not cohabiting; the only differences between the two groups were that those reporting no current dating relationship were more likely to be male and marginally less likely to be black. On the other hand, respondents who were not included because they were cohabiting or married differed from those who were dating in several ways: They were more likely to be female and Hispanic, and they were less likely to be black, to have lived with two biological or adoptive parents at baseline and to have a parent with at least some college experience. They also had a marginally lower average age at first sex. From these data, we created a relationship-level file, in which each record represented a relationship, for a total of 4,574. We then removed 19 relationships with missing data on contraceptive use at last sex and 16 relationships in which the respondent reported using methods that did not fit into our contraceptive method typology, described below. We also removed 386 relationships in which the respondents were not having sex with their dating partner. Although a respondent could have up to four records if he or she was currently dating a different partner at each interview date, more than 85% contributed just one relationship.* Because we could not look at multiple relationships for the majority of respondents, we focused on respondents’ most recent dating relationship and dropped 538 relationships that did not meet this criterion, leaving us with one relationship per respondent. Finally, we removed 130 relationships in which respondents said they were trying to get pregnant with their partner. Our final sample consisted of 3,485 unique relationships (1,804 from female respondents and 1,681 from male respondents).
Measures 䊉 Dependent variables. Our dependent variable of interest was a four-level measure of the type of contraceptive used at last sex: no method (including withdrawal and natural
*Respondents may have contributed only one relationship because they dated one partner across multiple rounds or because they had short relationships between survey dates that were no longer current at the time of the interview.
43
Contraceptive Use Within Young Adult Dating Relationships
family planning methods, which were reported by 2% of respondents); condom only; hormonal or long-acting method only; or dual method (condom and hormonal or long-acting method). 䊉 Relationship characteristics. We created six three-level categorical relationship variables for our latent class analysis. We chose three-level variables because having all variables on the same metric helps latent class analysis models converge31 and, in our data, the three-level categorical variables provided better model fit than two-level variables. We set our category cut points on the basis of previous research, but also to maintain a relatively even distribution across levels in our sample. Relationship duration was measured by the total number of months elapsed from the date the relationship began to the date of last sex. Responses were categorized as six months or less, seven months to one year or more than one year. Also, we included a variable measuring the length of the couple’s relationship prior to the first time they had sex (the presexual relationship). The categories indicated that the couple had had sex before or during the month the relationship started, had dated for 1–3 months prior to having sex or had dated for four or more months prior to having sex. Intimacy was assessed with two questions about how close respondents felt to their partner and how much they thought their partner cared about them. Responses for each ranged from 0 to 10 (high scores indicate high caring or closeness). Because these two variables were moderately correlated (r=0.66), and because a large proportion of respondents reported very high levels of closeness and caring, we averaged these measures and combined scores into one measure indicating low intimacy (a score of 7 or less), medium intimacy (a score of 8 or 9) or high intimacy (a score of 10). Relationship commitment was based on two measures. The first was a 0–10 scale in which respondents rated the likelihood that they and their partner would be together in six months. A score of zero indicated no chance; a 10 indicated 100% likelihood. We categorized this likelihood as low (less than 6), medium (6–9) or high (10). For the second measure of commitment, we included a variable for whether the couple had discussed marriage or cohabitation; responses were categorized as talked about neither, talked about cohabitation or talked about marriage. (Respondents who reported having discussed both cohabitation and marriage were included in the marriage category.) Respondents rated the level of conflict in their relationship on a 0–10 scale. Responses skewed toward the lower end of the scale, so we categorized them as low conflict *Eighteen respondents reported another race or ethnicity; these were included with whites. †We ran a latent class analysis model that included a three-level child variable (no child, child with current partner, child with other partner), and the results were unchanged.
44
(a score of 0–2), medium conflict (3–5) or high conflict (6–10). 䊉 Control variables. We included a number of family and individual controls associated with contraceptive use. Time-invariant characteristics measured at Round 1 were gender, race or ethnicity (white, black, native-born Hispanic and foreign-born Hispanic*), whether either of the respondent’s parents had completed some college or more, and whether the respondent had lived with two biological or adoptive parents at baseline. We also included a measure of the respondent’s age at first sex. Time-varying characteristics measured at the time of last sex with the dating partner were the respondent’s age, whether the respondent had completed some college or more, whether the respondent was neither enrolled in school nor employed, the number of sexual partners in the past year (capped at 10), whether the respondent had used hormonal methods prior to the current relationship and whether the respondent had children. Fifteen percent of the sample had had a child, but only 2% had had a child with the current partner. Therefore, for class creation, having had a child was considered an individual-level control, rather than a relationship-level characteristic.† Also, we controlled for several important partner characteristics. We included a measure of whether the partner was of a different race or ethnicity than the respondent, as well as a continuous measure of age difference between the respondent and his or her partner (where a positive value indicates the partner is older). Two variables measured whether the partner had completed some college or more and whether the partner was neither enrolled in school nor employed. Ideally, these two characteristics would be measured at the time of last sex; however, the relevant questions were based on the beginning of the relationship.
Analysis We conducted a latent class analysis in Mplus to identify relationship classes.31 We tested whether gender was a significant predictor of relationship class membership and found that it was not. Therefore, we conducted the class creation on the full sample of males and females. In conducting the latent class analysis, we compared the fit indices (entropy, the Bayesian information criterion, the sample-size adjusted Bayesian information criterion and the Akaike information criterion) for one- to five-class models to identify the appropriate number of classes. We then conducted descriptive and bivariate analyses to examine the average contraceptive patterns and partner and individual characteristics across classes, which we tested for differences using t tests. In multivariate analyses, we examined the association between relationship class and contraceptive use, net of partner, family and individual characteristics. Using logistic regression, we modeled contraceptive use versus nonuse (results not shown); using multinomial logistic regression, we modeled the four-category contraceptive method type. Regressions were run in Stata 11 and incorporated Perspectives on Sexual and Reproductive Health
household clustering and probability weighting variables. We ran the regression models four times, each time using a different relationship class as the reference group, in order to compare all of the classes. All associations presented were significant at p1 year
88.0 10.5 1.5
64.3 30.7 5.0
*,** 2.9 11.9 85.3
*,**,*** 16.1 15.6 68.3
Length of presexual relationship Sex before/in first month of dating 1–3 months ≥4 months
63.0 34.6 2.4
* 56.9 39.1 4.1
*,** 30.6 38.8 30.6
*,** 30.2 42.2 27.6
Intimacy† Low Medium High
77.6 22.1 0.3
* 0.7 70.5 28.7
*,** 29.1 50.1 20.7
*,**,*** 0.1 13.0 87.0
Likelihood of being with partner in six months‡ Low 76.9 Medium 21.8 High 1.3
* 8.1 70.1 21.8
*,** 38.1 55.2 6.7
*,**,*** 0.2 4.9 94.9
Discussed cohabitation/marriage Neither Cohabitation Marriage
72.8 18.0 9.3
* 38.5 21.2 40.2
*,** 18.5 12.1 69.4
*,**,*** 5.9 5.6 88.6
Conflict§ Low Medium High
39.8 36.6 23.6
* 57.8 18.5 23.7
*,** 15.6 45.8 38.6
*,*** 59.1 20.9 20.1
100.0
100.0
100.0
100.0
Total
*Differs from short-term/casual at p
